Tokyo System House Releases "AI Veteran Engineer" v1.2: Addressing COBOL Challenges with Implicit Knowledge Documentation and Claude Code Integration

Tokyo System House Co., Ltd. (Headquarters: Shinagawa-ku, Tokyo; Representative Director: Tomoyuki Hayashi; hereinafter "the Company"), a system integrator, announces the release of the latest version (v1.2) of "AI Veteran Engineer," which enables AI to handle COBOL specification document creation and Q&A, replacing the work of veteran engineers.

This version significantly improves the quality and stability of specification document generation, in addition to adding a feedback function to the Q&A system and strengthening integration with AI-driven development tools, including Claude Code. This will further streamline the maintenance and operation of COBOL systems and support them in a manner suitable for modern development styles.

Background

In the maintenance and operation of COBOL systems, knowledge and know-how accumulated over many years often become implicit and personalized, leading to time-consuming specification understanding and investigation.

Furthermore, while AI is increasingly used to improve development efficiency, AI-driven development tools alone face challenges in fully understanding vast COBOL codebases.

Against this backdrop, the Company has strengthened knowledge accumulation, utilization, and integration with AI-driven development in "AI Veteran Engineer." This will support COBOL system maintenance and operation in a practical manner.

Main Functions and Enhancements of "AI Veteran Engineer" v1.2

"AI Veteran Engineer" is a solution where AI replaces the tasks of veteran engineers proficient in COBOL systems. Version 1.2 adds the following functions to provide even more powerful support:

1. Added Feedback Function to Nurture the Q&A System

In the Q&A system, a function has been added that allows users to input descriptive feedback to AI Veteran Engineer's responses. This enables the accumulation of implicit knowledge and expert techniques held by veteran engineers, such as development precautions and special rules, as organizational knowledge through the feedback function.

Moreover, for issues that initially took a long time to investigate when building the Q&A system, early resolution becomes possible by registering chat history as "diagnosis patterns" in the knowledge base. As usage increases, knowledge accumulates, the system matures, and continuous accuracy improvement is achieved.

Feedback input screen for responses

2. Claude Code Integration

We have implemented integration with "Claude Code," an AI-driven development tool provided by Anthropic. Through MCP Server (SSE and Streamable-HTTP methods), AI Veteran Engineer can be called directly from Claude Code, allowing COBOL system knowledge to be utilized in AI-driven development.

Furthermore, by providing and setting COBOL expert skills, it is possible to maximize the extraction of knowledge held by AI Veteran Engineer.

Example of use with Claude Code

＜Benefits of utilizing AI Veteran Engineer's MCP Server in COBOL analysis＞

AI-driven development tools alone often struggle to fully understand vast COBOL codebases, leading to challenges such as fragmented specifications, inaccurate analysis due to diluted context, and increased costs. AI Veteran Engineer resolves these issues by pre-analyzing and vectorizing vast COBOL systems and then accurately providing necessary information to AI-driven development tools.

Benefits of utilizing AI Veteran Engineer's MCP Server

3. Improved Quality and Stability of Specification Document Generation

In specification document generation, "static analysis" handles deterministic processing, while "AI" handles semantic interpretation. This ensures the generation of consistent and highly accurate specification documents regardless of how many times they are output.

Furthermore, stable output is achieved by using placeholders and embedding values obtained by AI and static analysis tools into custom templates. This improves the quality of specification documents and further streamlines maintenance and operation.

4. Other Functional Improvements

・Improved visibility of specification documents

We have unified the expression of flowcharts and strictly enforced rules to achieve consistent output. Additionally, indentation has been added to the level number display of the WORKING-STORAGE SECTION, making the hierarchical relationship of data structures easier to visualize.

Example of flowchart output

Example of WORKING-STORAGE SECTION output

・Creation of program control structure diagrams

We have added a function to automatically generate a diagram of the overall program control structure within the specification documents. This allows for a visual understanding of the overall processing.

・Display of chat history on screen

In the Q&A system, chat history can now be viewed on screen.